Weight index of high-dimensional time-to-event feature selection method

Abstract Feature selection in high-dimensional data is challenging, aiming to eliminate irrelevant features and build a robust statistical model that identifies prognostic markers. Our proposed approach introduces a weight function that assigns relevance scores to each feature in the model. This weight function allows us to prioritize features in the dataset, enabling the selection of potential genes based on assigned weights and an updated threshold value. To highlight genes with the greatest variation, the coefficient of variation is integrated into the weight function. In addition, the model considers the proportion of selected factors throughout the iterative process to mitigate selection bias. We evaluate the proposed method using both training and testing datasets, producing a ranked list of filtered gene candidates. To demonstrate the method’s utility, we apply it to a gene expression dataset from the Gene Expression Omnibus, yielding a final selection of 12 features from an initial set of 507. All analyses are conducted in the R programming language. In summary, this method offers a novel approach to feature selection in high-dimensional data, particularly applicable to genomics, proteomics, and transcriptomics datasets used to predict and monitor disease progression. By effectively reducing selection bias, the method supports the creation of reliable statistical models that include valuable prognostic markers.